Presentation (CTI)

OBJECTIVES: Artificial intelligence (AI) is increasingly used in systematic literature reviews (SLRs); however, expectations for its responsible use vary by intended evidence application and stakeholder requirements. We aimed to develop a risk-based workflow to inform how to use AI responsibly and report transparently.
METHODS: We identified guidance through a targeted literature review (2021 to May 2026). Eligible documents included consensus guidance, reporting checklists, position statements, and recommendations on AI in evidence review, synthesis and HEOR evidence generation. We extracted, compared and synthesized findings across sources. We stratified our SLRs into three risk categories: low (using AI for internal or hypothesis-generating work), medium (using AI for publication-oriented evidence generation) and high (using AI for reviews informing comparative effectiveness analyses, quantitative analyses or decision making). We mapped the guidance to the workflow.
RESULTS: Our review identified important guidance, including the RAISE recommendations, the Cochrane/Campbell/JBI/CEE joint position statement, PRISMA-trAIce, the UK NICE position statement on AI in evidence generation, and ISPOR’s ELEVATE-GenAI framework. Comparative analysis showed convergence across five domains: requirements for transparency (reporting of AI use in the protocol and report), requirements for validation (obtaining and reporting performance measures from previous work and for the specific review), requirements for reproducibility (based on verifiable measures), requirements for human oversight (level of verification or decision making by subject matter experts for the specific review) and adherence to recognized methodology and reporting guidelines. We mapped guidance for each domain onto the SLR workflow for each of the three risk categories. This produced a stage-by-stage, quantifiable guide to using AI responsibly, with requirements for each domain increasing with risk level.
CONCLUSIONS: Our risk-based workflow offers a practical way to apply responsible AI in SLRs, meeting requirements for transparency, validation, reproducibility, human oversight and alignment with established standards for each category of evidence purpose.